Development of Data Driven Rainfall-Runoff Model for the Sarada River Basin

Abstract

Determining the relationship between rainfall and runoff for a basin is one of the challenging tasks faced by hydrologists and engineers. Conceptual rainfall-runoff models are most suitable in case of data scarcity. However, data driven models are more useful to handle nonlinearity between rainfall-runoff time series data. In this paper an attempt has been made to develop data driven models (Linear and non-linear models) for the Sarada river basin in Vishakhapatnam district of Andhra Pradesh, India. The catchment area of the Sarada river basin is 2665 Sq km. The observed daily rainfall obtained from IMD and daily runoff data obtained from CWC for a period of twenty four years (1989-2013). Autoregressive Integrated Moving Average (ARIMA) linear model and Artificial Neural Network (ANN) and Wavelet Neural Network (WNN) nonlinear models have been developed for the Sarada River basin. The 60% of observed data has been used for calibration and 40% of the data for validation. The comparison of model performance was conducted based upon different statistical indices. The result indicates WNN model performed better than ANN and AIRMA for rainfall-runoff modelling in the Sarada river basin.